Metrics for namespace "all"

Hi, Im looking into the new metrics for the new server version, Im seeing a namespace “all” in the metrics counters, can you help explain what this is?

also Im only seeing errors from that “all” namespace, our own namespace has no service errors

1 Like

Temporal emits a bunch of metric with namespace tag. We dual emit all such metric, one with the actual namespace and another for rollup with tag namespace=all.


The problem is that only namespace all has errors, all others doesnt…
Not sure where is that being accumulated from.
@samar

What is the metric name?

the metrics with values, is operation.namespace, which the namespace is “all”, other namespace has no errors so not sure how is this accumulated.

Prometheus has a restriction that it does not allow same metric to be emitted with different set of tags. Which means you cannot have a metric which in some places is emitted with namespace tag and other places without it. For this reason, if a metric which is emitted with a namespace tag even in a single location it requires us to have this tag in other places where it is not needed. So we decided to emit namespace tag with value all at the root scope level.
So in places where you see only all value for namespace then it means that metric is not emitted with a specific namespace.

“So in places where you see only all value for namespace then it means that metric is not emitted with a specific namespace.”
Sorry trying to understand this sentence, does that mean this is not a namespace related errors? Its an error happening in any of those namespaces?

Sorry trying to understand this sentence, does that mean this is not a namespace related errors? Its an error happening in any of those namespaces?

This just means that specific metric is not emitted with a namespace tag.

Ok this seems only applies to a few operations like AddActivityTask and etc, those operations is not associated with any namespace. Got it

will “not available” not be more appropriate than “all” for namespace?

I think ALL also makes sense as this gives you the rollup across all namespace for the metric.